Evidence chain-based continuous human-computer collaborative spatiotemporal semantic package generation method and system

Abstract

This invention belongs to the field of electronic data storage and spatiotemporal information processing, and discloses a method and system for generating spatiotemporal semantic packages based on continuous evidence chains through human-computer collaboration. The method involves collecting spatiotemporal data of the target business and recording information on the collection, transmission, processing, manual operation, and judgment processes. Hashes are generated for each item, and adjacent entries are hashed together to form a continuous evidence record sequence. Business events are extracted and atomically processed to construct a process graph containing temporal sequence, spatial correlation, processing dependencies, and manual operation relationships. The process graph representation learns to solve for the minimum evidence closure supporting the conclusion under constraints such as temporal continuity, dependency integrity, and operational traceability. The closure is encapsulated into a spatiotemporal semantic package, and after package hashing, timestamping, and digital signature, it is archived in a single set, achieving verifiable, traceable, and efficient evidence generation.

Description

Technical Field

[0001] This invention relates to the fields of electronic data storage and spatiotemporal information processing, and in particular to a method and system for generating spatiotemporal semantic packages based on human-computer collaboration with a continuous chain of evidence. Background Technology

[0002] With the widespread adoption of IoT sensing devices, mobile terminals, and industry information systems, spatiotemporal data containing information on time, location, object status, and processing results is continuously generated in business scenarios such as urban governance, public safety, traffic supervision, production safety monitoring, and asset inspection. Existing technologies typically employ a process of multi-source data collection, network transmission, platform processing, manual review or handling, and result archiving to complete the business loop. Key data is then documented or solidified through methods such as operation logs, processing logs, database auditing, electronic signatures, timestamps, and blockchain notarization. Simultaneously, some systems are beginning to introduce event extraction, process modeling, or knowledge graphs to structure and represent business processes, thereby improving retrieval and management efficiency.

[0003] However, in scenarios such as judicial evidence collection and audit verification, where the authenticity of evidence and the verifiability of its formation process are highly demanding, existing technologies still have the following shortcomings:

[0004] First, records from each stage are often scattered across different systems or data carriers of different formats, lacking a unified correlation mechanism and verifiable continuous links. This makes it difficult to form a traceable and undeniable continuous chain of evidence from the original sensory data, key processing steps, personnel operation traces, and final conclusions, which can easily lead to problems such as broken evidence chains or difficulty in proving the authenticity of the process.

[0005] Second, existing archives mainly consist of data files or result records, lacking a structured expression of human-computer collaboration processes, spatial semantic relationships, and business cognitive logic. This makes it difficult to atomize business elements and map them into single-set archive units, resulting in difficulties in achieving consistent preservation and review of business and archives throughout their entire lifecycle.

[0006] Third, when certificates or reviews are required, a large amount of data and logs usually need to be collected manually. There is a lack of a mechanism for generating the minimum necessary evidence set for the judgment conclusion, resulting in high evidence preparation costs, insufficient interpretability, and low certificate issuance efficiency.

[0007] Therefore, there is a need for a human-computer collaborative spatiotemporal semantic packet generation method and system that can overcome the shortcomings of the existing technologies. Summary of the Invention

[0008] One objective of this invention is to propose a method for generating spatiotemporal semantic packages based on a continuous chain of evidence for human-machine collaboration. This addresses the problem in existing technologies where there is a lack of consistent correlation mechanisms and verifiable process records among the various stages of spatiotemporal data acquisition, transmission, processing, judgment, and archiving. This results in the difficulty of forming a continuous chain of evidence from original perceived data, key processing steps, human operation traces, and the final judgment conclusion. Furthermore, archiving primarily consists of data files or result records, making it difficult to structurally express the human-machine collaboration process and spatial semantics, and to achieve single-set preservation and verification. The invention proposes the following technical solution: Acquire spatiotemporal data of the target business process and synchronously record process information; [The method then continues with further details about the generation of spatiotemporal semantic packages based on a continuous chain of evidence for human-machine collaboration, including the generation of spatiotemporal semantic packages from original perceived data, transmission, processing, judgment, and archiving.] Process information, manual operation records, and judgment conclusions are each generated with hash digests and linked by hash association between adjacent entries to form a continuous evidence record sequence. Business event extraction and atomization are performed on the evidence data, and event atoms are associated with evidence entry identifiers and hash digests. A process graph containing temporal sequence, spatial association, processing dependencies, and manual operation relationships is constructed and its representation is learned. Under the proof constraints of temporal continuity, processing dependency integrity, and traceability of manual operations, the minimum evidence closure supporting the judgment conclusion is solved. The minimum evidence closure is encapsulated into a spatiotemporal semantic package, and after package hashing, timestamping, and digital signature, it is stored as a single-set archiving unit. This invention possesses the technical advantages of a continuous and verifiable evidence chain, a traceable and non-repudiable formation process, a reproducible and verifiable archiving unit, and high certification efficiency.

[0009] On the one hand, this invention provides a method for generating spatiotemporal semantic packets through human-computer collaboration based on a continuous chain of evidence, including:

[0010] S1. Collect spatiotemporal data corresponding to the target business process, and synchronously record process information of the spatiotemporal data during collection, transmission, processing, and judgment to form an evidence dataset. Generate hash digests for the original perceived data, process information, manual operation records, and judgment conclusions in the evidence dataset. Generate evidence record entries in chronological order, and form a continuously linked sequence of evidence record entries based on the hash association between adjacent evidence record entries. S2. Based on the evidence dataset and the sequence of evidence record entries, perform business event extraction and atomization processing on the original perceived data, process information, manual operation records, and judgment conclusions to generate a set of business event atoms. Identify the evidence record entries corresponding to each business event atom association and assign them their hash digests. S3. Construct a process diagram representing the human-machine collaboration process based on the set of business event atoms and the sequence of evidence record entries, including business events. The process graph consists of atomic nodes and evidence record entry nodes, as well as edges representing temporal sequence, spatial correlation, processing dependency, and manual operation relationships; S4, performing process graph representation learning on the process graph to generate a graph representation feature set, tracing and solving the atomic nodes of business events corresponding to the judgment conclusion, and selecting the minimum number of nodes and edges required to support the judgment conclusion from the process graph under the condition of satisfying the preset proof constraints, generating a minimum evidence closure. The preset proof constraints include at least temporal continuity constraints, processing dependency integrity constraints, and manual operation traceability constraints; S5, encapsulating the minimum evidence closure to generate a spatiotemporal semantic package, and archiving the spatiotemporal semantic package as a single-set archiving unit, generating a package hash digest for the spatiotemporal semantic package, generating a timestamp for the package hash digest, and storing it after digital signature to support the integrity verification and traceability of the formation process of the spatiotemporal semantic package.

[0011] Optionally, S1 includes:

[0012] Collect raw sensing data corresponding to the target business process, and simultaneously generate time information and location information corresponding to the raw sensing data during the collection process;

[0013] The process information of the original sensing data during the acquisition, transmission, processing and judgment process is recorded synchronously. The process information includes at least the acquisition device identification information, acquisition parameter information, transmission link identification information, processing program identification information, processing parameter information and operator identification information.

[0014] The original sensing data, the process information, the manual operation record, and the judgment conclusion are standardized and encoded respectively, and then hash digests are calculated. An incremental sequence number is assigned to the hash digest corresponding to each item.

[0015] Evidence record entries are generated in the order of the incrementing sequence number. In each evidence record entry, the incrementing sequence number of the evidence record entry, the hash digest of the evidence record entry, the time information corresponding to the evidence record entry, the location information corresponding to the evidence record entry, and the hash digest of the previous evidence record entry are written.

[0016] Each evidence record entry is appended and stored in the order of its generation to form the continuously linked evidence record entry sequence, and the original perception data, the process information, the manual operation records, and the judgment conclusions, together with their corresponding hash digests, are collected to form the evidence dataset;

[0017] Furthermore, the hash digests corresponding to the evidence record entries generated within the preset time window are constructed into a Merkle tree, and the Merkle root hash is sent as an anchor digest to a third-party timestamp service or distributed ledger for evidence storage, and the returned evidence storage receipt is written into the evidence record entry sequence.

[0018] Optionally, S2 includes:

[0019] Based on the time and location information carried in the evidence dataset, the original perception data, process information, manual operation records and judgment conclusions are time-aligned and spatially aligned to form candidate event fragments arranged in chronological order.

[0020] For each candidate event fragment, perform event element extraction to obtain the event time, event location, event subject, event action, and event object corresponding to the candidate event fragment;

[0021] The event elements are atomically segmented according to the preset event template to generate business event atoms, and a unique event identifier is assigned to each business event atom.

[0022] Based on the event time and event location of each business event atom, a matching evidence record entry is retrieved in the evidence record entry sequence. The incrementing sequence number of the retrieved evidence record entry is determined as the evidence record entry identifier of the business event atom, and the hash digest of the evidence record entry is determined as the hash digest of the business event atom.

[0023] All business event atoms are aggregated to form a business event atom set, where each business event atom includes at least an event identifier, event time, event location, event subject, event action, event object, corresponding evidence record entry identifier, and corresponding hash digest.

[0024] Optionally, S3 includes:

[0025] Each business event atom in the business event atom set is constructed as a business event atom node, and each evidence record entry in the evidence record entry sequence is constructed as an evidence record entry node, thus forming a node set of the process diagram;

[0026] The business event atomic nodes are sorted according to the event time in the business event atomic set, and time sequence edges are established between adjacent business event atomic nodes in the sort.

[0027] Based on the event location in the business event atomic set, establish spatial association edges for business event atomic nodes that meet the spatial adjacency conditions according to preset spatial adjacency rules;

[0028] Based on the process information in the evidence dataset, identify the calling relationship or data dependency relationship between the data processing records corresponding to the atomic business events, and establish processing dependency relationship edges for the atomic nodes of business events that have calling relationships or data dependencies.

[0029] Based on the manual operation records in the evidence dataset, identify the association between the manual operation records and the corresponding business event atoms, and establish a manual operation relationship edge between the evidence record entry node corresponding to the manual operation record and the corresponding business event atom node.

[0030] Write the relationship type field for the time sequence relationship edge, spatial relationship relationship edge, processing dependency relationship edge and manual operation relationship edge respectively, and write the edge weight for each edge according to the preset weighting rule to obtain the process diagram representing the human-computer collaboration process.

[0031] Optionally, S4 includes:

[0032] Write node attributes for the business event atomic nodes in the process diagram. The node attributes include at least event time, event location, event subject, event action, event object, evidence record entry identifier, and hash digest. Write node attributes for the evidence record entry nodes in the process diagram. The node attributes include at least incrementing sequence number, time information, location information, and hash digest.

[0033] Input the node attributes and the relationship type field and edge weight of each edge in the process graph into the process graph representation learning model to obtain the graph representation feature set corresponding to each node and each edge;

[0034] Based on the feature set of the graph representation, the evidence contribution degree of the atomic node of the business event corresponding to the judgment conclusion is calculated, and the evidence contribution degree is used as the search weight to perform a retrospective search in the process graph to obtain a candidate closed subgraph containing the atomic node of the business event corresponding to the judgment conclusion.

[0035] In the candidate closure subgraph, nodes and edges are removed in order of their evidence contribution from low to high. After each removal, the preset proof constraints are verified. If the verification passes, the removal result is retained. If the verification fails, the removal result is rolled back until there are no nodes or edges in the candidate closure subgraph that can be removed and still satisfy the preset proof constraints, thus obtaining the minimum evidence closure.

[0036] The preset proof constraints include at least a time continuity constraint, a processing dependency integrity constraint, and a manual operation traceability constraint. The time continuity constraint is used to limit the atomic nodes of the business events corresponding to the judgment conclusion in the minimum evidence closure to be traced along the time sequence edge to the preceding atomic nodes of the business events with corresponding evidence record entries. The processing dependency integrity constraint is used to limit the minimum evidence closure to include all preceding atomic nodes of the business events corresponding to the judgment conclusion that have processing dependency edges. The manual operation traceability constraint is used to limit the minimum evidence closure to include evidence record entries that have manual operation edges of the business events corresponding to the judgment conclusion.

[0037] Furthermore, the process graph representation learning model is a heterogeneous relational graph neural network model, which uses different relational parameters for message passing on the temporal sequence relational edges, spatial relational edges, processing dependency relational edges, and manual operation relational edges, and introduces time position encoding for event time to generate the graph representation feature set;

[0038] Furthermore, the preset proof constraint further includes a process reproducibility constraint, which is used to limit the minimum evidence closure to include at least the process identification information, process version information, process parameter information, and runtime environment fingerprint information related to the judgment conclusion; wherein, the runtime environment fingerprint information includes at least one of container image digest, runtime library version digest, or execution environment proof digest.

[0039] Optionally, S5 includes:

[0040] Based on the minimum evidence closure, the original perceived data reference information, process information, manual operation records and judgment conclusions corresponding to the business event atomic nodes and evidence record entry nodes contained in the minimum evidence closure are extracted and summarized to obtain the spatiotemporal semantic package content.

[0041] The content of the spatiotemporal semantic package is structured and encapsulated to generate a spatiotemporal semantic package. The spatiotemporal semantic package includes at least a spatiotemporal semantic package identifier, a judgment conclusion, a set of business event atoms associated with the judgment conclusion, a sequence of evidence record entries associated with the set of business event atoms, original perception data reference information, process information, manual operation records, time information, and location information.

[0042] The original sensory data reference information, the evidence record entry sequence, the process information, the manual operation record, and the judgment conclusion in the spatiotemporal semantic package are concatenated in a preset order and a packet hash digest is calculated.

[0043] The packet hash digest is generated with a timestamp, digitally signed, and then stored.

[0044] The spatiotemporal semantic package, the package hash digest, the timestamp, and the digital signature are written into the archive storage as a single-set archive unit to support the integrity verification of the spatiotemporal semantic package and the traceability of its formation process.

[0045] On the other hand, the present invention also provides a human-computer collaborative spatiotemporal semantic packet generation system based on a continuous chain of evidence, including:

[0046] The evidence chain generation module is used to collect spatiotemporal data of the target business process and record the corresponding process information. It generates hash digests of the original perception data, process information, manual operation records and judgment conclusions, generates evidence record entries in chronological order and forms a continuous evidence record entry sequence by hash association between adjacent entries.

[0047] The event atomization module is used to extract and atomize business events based on the evidence data and the sequence of evidence record entries, generate a set of business event atoms and associate each business event atom with the corresponding evidence record entry identifier and its hash digest.

[0048] The process diagram construction module is used to construct process diagrams using business event atomic nodes and evidence record entry nodes, and to establish edges for temporal sequence, spatial association, processing dependency, and manual operation relationships.

[0049] The minimum evidence closure generation module is used to perform representation learning on the process graph and trace back to solve the atomic nodes of the business events corresponding to the judgment conclusion. Under the conditions of satisfying the constraints of time continuity, processing dependency integrity and traceability of manual operation, the minimum evidence closure supporting the judgment conclusion is generated.

[0050] The semantic package archiving module is used to encapsulate the minimum evidence closure into a spatiotemporal semantic package and archive it as a single-set archiving unit. It also generates a package hash digest for the spatiotemporal semantic package and stores it with a timestamp and digital signature to support integrity verification and traceability of the formation process.

[0051] The beneficial effects of this invention are:

[0052] 1. By generating hash digests from the original perception data, process information, manual operation records and judgment conclusions respectively, and forming a continuous evidence record sequence by hash association of adjacent evidence record entries, combined with packet-level hash, timestamp and digital signature storage, the integrity of the data and the formation process can be verified, traceable and non-repudiable, reducing the risk of evidence chain breakage.

[0053] 2. By extracting and atomizing business events, key business elements in the human-machine collaboration process are structured into business event atoms. Event atoms are then bound to evidence record entries and hash digests. This further constructs a process diagram that includes temporal sequence, spatial association, processing dependencies, and relationships with manual operations. This achieves a consistent mapping between business semantics and the evidence chain, enhancing the ability to interpret and verify evidence.

[0054] 3. By using process diagrams to represent learning and proof constraint verification, the system automatically solves the minimum evidence closure that supports the judgment conclusion under the conditions of time continuity, processing dependency integrity and traceability of manual operation, and encapsulates it into a spatiotemporal semantic package as a single-set archiving unit, thereby reducing the scope of evidence required for certification and the cost of manual sorting, and improving archiving efficiency and certification efficiency. Attached Figure Description

[0055] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0056] Figure 1 This is a flowchart of the human-computer collaborative spatiotemporal semantic packet generation method based on continuous evidence chains proposed in this invention. Detailed Implementation

[0057] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0058] refer to Figure 1 A human-computer collaborative spatiotemporal semantic packet generation method based on continuous evidence chains includes:

[0059] S1. Collect spatiotemporal data corresponding to the target business process, and synchronously record process information of the spatiotemporal data during collection, transmission, processing, and judgment to form an evidence dataset. Generate hash digests for the original perceived data, process information, manual operation records, and judgment conclusions in the evidence dataset. Generate evidence record entries in chronological order, and form a continuously linked sequence of evidence record entries based on the hash association between adjacent evidence record entries. S2. Based on the evidence dataset and the sequence of evidence record entries, perform business event extraction and atomization processing on the original perceived data, process information, manual operation records, and judgment conclusions to generate a set of business event atoms. Identify the evidence record entries corresponding to each business event atom association and assign them their hash digests. S3. Construct a process diagram representing the human-machine collaboration process based on the set of business event atoms and the sequence of evidence record entries, including business events. The process graph consists of atomic nodes and evidence record entry nodes, as well as edges representing temporal sequence, spatial correlation, processing dependency, and manual operation relationships; S4, performing process graph representation learning on the process graph to generate a graph representation feature set, tracing and solving the atomic nodes of business events corresponding to the judgment conclusion, and selecting the minimum number of nodes and edges required to support the judgment conclusion from the process graph under the condition of satisfying the preset proof constraints, generating a minimum evidence closure. The preset proof constraints include at least temporal continuity constraints, processing dependency integrity constraints, and manual operation traceability constraints; S5, encapsulating the minimum evidence closure to generate a spatiotemporal semantic package, and archiving the spatiotemporal semantic package as a single-set archiving unit, generating a package hash digest for the spatiotemporal semantic package, generating a timestamp for the package hash digest, and storing it after digital signature to support the integrity verification and traceability of the formation process of the spatiotemporal semantic package.

[0060] In this specific embodiment, S1 includes:

[0061] The data acquisition terminal collects raw sensing data corresponding to the target business process and obtains time information from the time synchronization source at the moment of acquisition. And obtain location information from the location source Time information Location information is represented by a millisecond-level timestamp based on UTC and written by the acquisition terminal. This indicates that the longitude, latitude, and altitude triplets are used in the WGS-84 coordinate system and are written by the acquisition terminal.

[0062] Simultaneously, the acquisition terminal and the business processing system generate process information records at each stage of data acquisition, transmission, processing, and judgment. The process information records include at least the acquisition device identification information, acquisition parameter information, transmission link identification information, processing program identification information, processing parameter information, and operator identification information. Among them, the acquisition device identification information is a unique device identifier string that is permanently bound to the acquisition terminal. The acquisition parameter information includes the sampling frequency, resolution, and sensor working mode and is recorded in the form of field-based key-value pairs. The transmission link identification information includes the system identifiers at both ends of the link, the transmission protocol identifier, and the session identifier, and is written by the network stack when establishing the connection. The processing program identification information includes the processing program name, processing program version number, and processing program binary hash, and is written by the business processing system when loading the program. The processing parameter information includes the threshold, model version identifier, and policy switch, and is written by the business processing system when the task starts. The operator identification information includes the operator account identifier and terminal login session identifier, and is written by the identity authentication system.

[0063] Subsequently, the raw sensing data, process information records, manual operation records, and judgment conclusions were converted into standardized coding results. The structured records adopt the standardized JSON serialization rules of UTF-8 encoding and meet the constraints that field names are sorted in lexicographical order, values ​​are represented in decimal without exponent, and time information fields are uniformly written as millisecond-level timestamp strings without whitespace characters. The binary raw perception data adopts the concatenation encoding rule of "fixed type header + length header + raw byte stream". The type header is taken from the media type identifier string preset by the acquisition terminal, and the length header is taken from the decimal string of the length of the raw byte stream and concatenated with the raw byte stream in order to eliminate ambiguity.

[0064] For each normalized coding result Calculate hash digest And assign an increasing sequence number Incrementing sequence number An atomic counter, maintained by the evidence chain generation module, is used to generate and increment sequentially from 1 to ensure global uniqueness and consistency with the time order. The hash digest is then used. According to the formula Calculation, where Indicates the first Each data item corresponds to a hash digest, and its length is 32 bytes. This refers to a cryptographic hash function identified by the SHA-256 algorithm. Indicates the first The normalized encoding result byte sequence of each data item Indicates the sequential index of evidence record entries and is associated with an incrementing sequence number. One-to-one correspondence;

[0065] When generating evidence record entries, the serial number will be incremented. Hash digest of this entry The time information corresponding to this entry Location information corresponding to this entry and the hash digest of the previous evidence record entry Write the same evidence record entry and append it to the sequence of evidence record entries, where A fixed 32-byte all-zero value is used as the chain head preorder digest and for verifying the chain start point. The evidence record entry sequence is stored as an append-only, non-overwrite data structure with an incrementing sequence number. This serves as the physical write order and retrieval primary key to ensure that the hash association between adjacent entries can be recalculated and verified.

[0066] While the sequence of evidence record entries is continuously generated, the system operates within a fixed time window. The hash digests of evidence record entries falling into the same window are aggregated and a Merkle tree is constructed, where the time window... Indicates the anchoring period and the window is aligned with the UTC whole minute boundary; window index. Indicates the first A time window is defined by the start time of each window. The leaf set takes the hash digest sequence sorted by ascending index within that window and combines adjacent leaf hash digests pairwise according to the byte concatenation rule of "left value first, right value last". Then, SHA256(•) is performed again to obtain the hash digest of the node in the next layer. When the number of nodes in a certain layer is odd, the hash digest of the last node is copied and concatenated with itself to form a pair of inputs, and the process continues to iterate upwards until a unique Merkle hash is obtained. ,in Indicates the first An anchored digest for each time window and a commitment value for the set of hash digests of all evidence record entries in that window;

[0067] The system will use Merkle root hashes The message imprint field is submitted via HTTPS by calling a third-party timestamp service and following the RFC3161 timestamp request format. Furthermore, the hash algorithm is identified as SHA-256, and the third-party timestamp service returns a timestamp receipt. ,in This represents a timestamp token byte sequence containing the timestamp time, the service provider's certificate chain, and the signature value;

[0068] The system will issue a receipt. Along with window index The window's start and end times, the range of incrementing sequence numbers covered by the window, and the Merkle root hash. The organization generates a corresponding anchored evidence receipt record and uses the same standardized coding rules as described above. with incrementing sequence number Then, the evidence record entry corresponding to the anchored evidence receipt is added to the evidence record entry sequence, so that the third-party anchoring information becomes part of the continuous evidence chain. The original perception data, process information records, manual operation records, judgment conclusions, anchored evidence receipt records and their corresponding hash digests and incrementing sequence numbers are collected to form an evidence dataset for subsequent event atomization and process diagram construction.

[0069] In this specific embodiment, S2 includes:

[0070] The event atomization module reads the incrementing sequence number of each evidence record from the evidence dataset and the sequentially linked sequence of evidence record entries. Time information Location information Hash digest And the data type identifier associated with the evidence record entry and the establishment of time-based information. An ascending index table is used as the time alignment reference, where This indicates the sequential index of the evidence record entry within the sequence of evidence record entries, and is related to the incrementing sequence number. One-to-one correspondence, time information Location information The meaning is the same as step S1;

[0071] The event atomication module aligns windows with fixed time. Alignment radius with fixed space Perform temporal and spatial alignment and generate candidate event fragments, specifically by using each evidence record entry as an anchor entry and its temporal information. As the start time of the segment, the retrieval satisfies And the spherical distance between the position information of the two entries is no greater than All evidence records are grouped into the same candidate event segment in ascending order of chronological information. This represents the index of retrieved evidence records. The spherical distance is calculated using the WGS-84 geodesic distance algorithm and takes the Earth's semi-major axis. As a calculation parameter to ensure that the distance calculation is definite and can be recalculated;

[0072] For each candidate event fragment, perform event element extraction and obtain the event time. Event location Subject of the event Event Actions With event object Event time The time information and event location of the evidence record entry with the minimum time information among the candidate event segments are selected. Get Event Time The location information of the corresponding evidence record item, and the subject of the event. The event action is determined according to the rule of "human intervention first, equipment as a backup". If a human operation record exists in the candidate event segment, the operator identification information in the human operation record is used; otherwise, the data acquisition device identification information in the process information is used. The mapping from the record type within a candidate event fragment to the action dictionary is determined, and this action dictionary permanently includes five types of action identifiers: acquisition, transmission, processing, verification, and judgment. These correspond to the occurrence or absence of original sensor data, transmission link identifier information, processing program identifier information, manual operation records, and judgment conclusions, respectively, and the event object. Take the pre-set business object identifier field from the process information record within the candidate event segment, and when the field is missing, take the object identifier field from the normalized coding result of the original perception data to ensure that the object field is unique and traceable.

[0073] Subsequently, the event elements are atomically segmented according to the preset event template to generate business event atoms, and a unique event identifier is assigned to each business event atom. The preset event template is fixed as a structure containing five fields: event time, event location, event subject, event action, and event object, and the values ​​of these fields are fixed. Event identifier The SHA-256 is calculated from the byte sequence of the string "EVT" and the event element field in a fixed order, and the first 16 bytes are truncated and then encoded in hexadecimal to ensure that the generation is consistent and non-repeating across systems.

[0074] Based on the event time of each business event atom Location of the event Retrieve matching evidence record entries from the sequence of evidence record entries and complete the evidence binding, specifically by constructing a candidate index set. And select the matching index within that set. Make ,in This represents the set of indexes of evidence records that satisfy both temporal and spatial alignment constraints. This represents the index of the final matched evidence record entries. This represents the event time of a business event atom. Indicates that the index is The time information of the evidence record entries Indicates the event position of the business event atom. Indicates that the index is Location information of evidence record entries, This represents the spherical distance calculated using the above geodesic distance algorithm, with units of 1. This represents a fixed weight that converts spatial distance into time cost in order to achieve a unified cost measurement for both time and space.

[0075] index The incrementing sequence number of the corresponding evidence record item Identify the evidence record entry that is the atom of the business event and generate a hash digest of the evidence record entry. The hash digest of the atomic business event is determined so that each atomic business event contains at least an event identifier. Event Time Event location Subject of the event Event Actions Event object Evidence record item identifier and hash digest And all business event atoms are aggregated to form a business event atom set.

[0076] In this specific embodiment, S3 includes:

[0077] The process diagram construction module constructs a process diagram representing the human-machine collaboration process based on the business event atom set and the evidence record item sequence. The data structure of the process diagram follows... Definition, where Representing the process diagram, The set of nodes representing the process diagram. Represents the set of edges in a process graph. This represents a set of atomic nodes for business events, with each node identified by an event identifier. It serves as the primary key for the node and includes the event time, event location, event subject, event action, event object, and the identifier of the evidence record entry. With hash digest This represents a set of evidence record nodes, with each node numbered in ascending order. Use it as the node's primary key and write time information. Location information With hash digest ,in Meaning and steps and steps Consistent;

[0078] The process graph construction module first instantiates each business event in the business event atom set into a business event atomic node and adds it to the process graph construction module. Simultaneously, each evidence record entry in the evidence record entry sequence is instantiated as an evidence record entry node and added to the list. This completes the initialization of the node set;

[0079] Subsequently, time sequence edges are constructed based on the temporal order. Specifically, all atomic nodes of business events are sorted in ascending order of event time, and in the case of identical event times, they are sorted by event identifier. The lexicographical order is used to achieve a stable sort. After sorting, a directed edge is written between two adjacent atomic nodes of business events, with the direction from the previous node to the next node. The relation type field TEMPORAL is written on this edge, and the edge weight is 1.0 to indicate the determinism of the time sequence.

[0080] Subsequently, spatial relationship edges are constructed based on spatial relationships, specifically in the following steps. Spatial alignment radius As a spatial adjacency threshold and in steps The geodesic distance algorithm used in the process is used as the distance calculation rule. For each business event atomic node, all other business event atomic nodes whose event positions are within the spatial adjacency threshold are retrieved. For each pair of nodes that meet the spatial adjacency condition, an undirected edge is written and stored with two directed edges in opposite directions. The relationship type field SPATIAL is written on the edge and the edge weight is 0.7 to represent the spatial association relationship.

[0081] Subsequently, processing dependency edges are constructed based on processing dependencies. Specifically, the process information records corresponding to the atomic nodes of business events are read from the evidence dataset, and the input increment sequence number list field and output increment sequence number field are parsed. The input increment sequence number list field represents the set of increment sequence numbers of upstream evidence record entries consumed by this processing step and is written by the processing program when reading input data. The output increment sequence number field represents the increment sequence number of output evidence record entries generated by this processing step and is written by the processing program when writing output data. For each input increment sequence number, the process graph construction module retrieves the previous business event atomic node whose evidence record entry identifier is equal to the input increment sequence number in the business event atomic set, and writes a directed edge from the previous business event atomic node to the current business event atomic node. The relationship type field is set to DEPENDENCY and the edge weight is set to 1.3 to represent the strong constraint of the processing dependency.

[0082] Subsequently, a manual operation relationship edge is constructed based on the manual operation relationship. Specifically, the evidence record entry node corresponding to the manual operation record is read from the evidence dataset, and the operated incremented sequence number field in the manual operation record is parsed. The operated incremented sequence number field represents the incremented sequence number of the evidence record entry corresponding to the data to be reviewed or the task to be handled selected by the operator in the business interface, and is written by the business system when submitting the manual operation. The process graph construction module uses the operated incremented sequence number as the key to search for the business event atomic node in the business event atomic set whose evidence record entry identifier is equal to the operated incremented sequence number. A directed edge is written between the evidence record entry node corresponding to the manual operation record and the retrieved business event atomic node, with the direction from the evidence record entry node to the business event atomic node. The relationship type field is written to this edge with the value MANUAL, and the edge weight is written with the value 1.5 to represent the intervention relationship of the manual operation on the business event.

[0083] The process graph construction module adds the above four types of edges to the edge set. It ensures that each edge contains the source node primary key, the target node primary key, the relationship type field, and the edge weight field, thereby obtaining and outputting a process graph that includes temporal relationships, spatial relationships, processing dependencies, and manual operation relationships.

[0084] In this specific embodiment, S4 includes:

[0085] Minimum Evidence Closure Generation Module Receiving Process Diagram The node and edge attributes are then permanently written to form a computable input, with the event time written to the atomic node of each business event. Event location Subject of the event Event Actions Event object Evidence record item identifier and hash digest Write an incrementing sequence number to each evidence record entry node. Time information Location information and hash digest For each edge, a relation type field and an edge weight field are written, with the relation type field only taking the values ​​TEMPORAL, SPATIAL, DEPENDENCY, and MANUAL, and the edge weights being fixed values. With 1.5 and steps Consistent;

[0086] The minimum evidence closure generation module inputs the aforementioned node attributes and edge attributes into the heterogeneous relational graph neural network model. The execution process involves learning graph representations and outputting a graph representation feature set, where the model... Depend on Layered relation-aware message passing network structure and hidden dimension selection And configure an independent relation parameter matrix for each relation type at each level. Where r represents the relation type and is taken from {TEMPORAL, SPATIAL, DEPENDENCY, MANUAL}, The layer index is represented and its value is 1 or 2;

[0087] Model The event time is encoded using time-location coding, and the event location is encoded using spatial location coding. The time-location coding incorporates the event time... Discretize the time slot index with seconds as the step size and embed the time in a time table of fixed size 4096 and dimension 16. The time vector is obtained by looking up a table, and the spatial location is encoded to represent the event location. according to The grid is divided into a grid index and embedded in a table in a space of fixed size 65536 and dimension 16. The spatial vector is obtained by looking up a table, and then the event subject is... Event Actions With event object The determined string hash modulo is mapped to an index, and the vector is obtained by looking up the main embedding table, action embedding table, and object embedding table. The above vector is then concatenated with the node type embedding vector and then passed through the input projection matrix. Linear transformation yields the initial node representation ,in Represents a node The initial representation vector, This indicates the dimension of the concatenated vector, which is determined by the sum of the dimensions of the embedded vectors;

[0088] In each layer In message passing, the model Apply the corresponding method for different relation types. A linear transformation is performed on the neighbor node representation, and the neighbor messages are weighted and summed using the edge weight field to achieve relational differentiation and evidence strength injection. ReLU activation and layer normalization are used between layers, and the dropout rate is fixed at 0.1 to suppress overfitting.

[0089] Model During the offline training phase, parameter learning is performed using the "business event atomic nodes corresponding to the judgment conclusions" in the historical process graph as the supervised target. The training batch size is 32, the number of training epochs is 20, the optimizer is Adam, and the learning rate is [missing value]. And fix the weight decay coefficient as After training, all parameters are frozen for use in subsequent steps. Reasoning and calculation;

[0090] The minimum evidence closure generation module determines the atomic node of the business event corresponding to the judgment conclusion as the conclusion node. And represented by the final node output by the model. Calculate each node Evidence contribution The contribution of evidence is used as the sole weight for retrospective search and elimination ranking. According to the formula:

[0091] ;

[0092] in Represents a node Relative to the conclusion node Evidence contribution and its range is Represents a node The final node represents a vector. Represents the conclusion node The final node represents a vector. This represents the transpose of a vector. Represents the L2 norm;

[0093] The minimum evidence closure generation module is based on Prioritize the search from the conclusion node The process begins with a traceback search to generate candidate closure subgraphs. During the search process, each included business event atomic node is simultaneously included with its evidence record entry identifier. The corresponding evidence record entries are used to ensure a consistent mapping between events and evidence record entries. All preceding business event atomic nodes and their corresponding evidence record entries are recursively included along the DEPENDENCY edge. Preceding business event atomic nodes traceable from the conclusion node are continuously included along the TEMPORAL edge until no preceding TEMPORAL edge exists. Evidence record entries corresponding to all manual operation records associated with the included business event atomic nodes via the MANUAL edge are also included to form a candidate closed subgraph containing the conclusion support path. ;

[0094] In obtaining Subsequently, the minimum evidence closure generation module uses the node evidence contribution degree Together with the edge weight field, the edge removal order is determined, and an iterative minimization process of "removing edges one by one from low to high contribution, verifying them after removal, and backing up if verification fails" is executed to obtain the minimum evidence closure. Each removal is simultaneously updated and Furthermore, each verification performs a full check on the preset proof constraints, which include time continuity constraints, processing dependency integrity constraints, manual operation traceability constraints, and processing reproducibility constraints.

[0095] The validation rule for time continuity constraints is to start from the conclusion node in the current subgraph. When backtracking along the TEMPORAL edge, each step has a corresponding preceding business event atomic node, and the evidence record entry of that preceding business event atomic node is identified. The corresponding evidence record entry node also exists in the current subgraph;

[0096] The validation rule for handling dependency integrity constraints is to check each atomic node of a business event in the current subgraph if it exists in the original process graph. If there is a preceding business event atomic node with an incoming edge relationship type of DEPENDENCY, then all preceding business event atomic nodes and their corresponding evidence record entries exist in the current subgraph.

[0097] The verification rule for the traceability constraint of manual operation is to check the conclusion node. and the set of business event atomic nodes included through the DEPENDENCY and TEMPORAL constraints, all in the original process diagram When there is an edge pointing to the atomic node of the business event and the relation type is MANUAL, the source evidence record entry node of the edge must exist in the current subgraph;

[0098] The validation rule for handling reproducibility constraints is to apply the event actions in the current subgraph. For the atomic nodes of business events that are processed or judged, their corresponding evidence record entries are identified. The evidence record entry node must contain the processor identifier, processor version, processing parameters, and runtime environment fingerprint information. Among them, the operating environment fingerprint information The evidence record entry node attribute is written with at least one of the following as input: container image digest, runtime version digest, and execution environment proof digest, and is written to the evidence record entry node attribute when the process information record is generated to ensure that the processing environment can be reconstructed during review;

[0099] When the iteration process finds no nodes or edges that can be further eliminated while still satisfying all the preset proof constraints, the current subgraph is determined to be the minimum evidence closure. It also outputs data for spatiotemporal semantic package encapsulation and archiving.

[0100] In this specific embodiment, S5 includes:

[0101] The semantic packet archiving module receives the minimum evidence closure. The set of business event atomic nodes and the set of evidence record entry nodes contained in the closure are parsed out. Then, the evidence record entry nodes are sorted in ascending order by their incrementing sequence number to obtain the sequence of evidence record entries associated with the judgment conclusion, and the content extraction and encapsulation are performed accordingly.

[0102] During content extraction, the semantic package archiving module reads the event identifier from the atomic node of each business event. Event Time Event location Subject of the event Event Actions Event object Evidence record item identifier With hash digest Read the incrementing sequence number for each evidence record entry node. Time information Location information With hash digest Based on the data type identifier recorded in the evidence record entry node, the corresponding original perception data, process information, manual operation records and judgment conclusions are retrieved from the evidence dataset.

[0103] Instead of directly embedding the raw sensing data into the spatiotemporal semantic package, raw sensing data reference information is generated and written into the spatiotemporal semantic package. This raw sensing data reference information consistently includes an object storage address field, an object storage bucket field, an object key field, an object length field, and an object content hash field. The object content hash field is obtained through specific steps. The hash digest generated from the raw sensory data ensures that the referenced target is verifiable and consistent with the chain of evidence; process information and manual operation records are all processed step by step. The standardized encoding rules are written into a field-based key-value structure, retaining the acquisition device identification information, acquisition parameter information, transmission link identification information, processing program identification information, processing program version information, processing parameter information, operator identification information, and operating environment fingerprint information. Fields used for review and reproduction, and the judgment conclusions are fixed to include the conclusion type field, conclusion value field, conclusion generation time field, and the event identifier field of the atomic node of the business event corresponding to the conclusion, so as to ensure the definite binding of the conclusion and the starting point of the closure;

[0104] During the structured encapsulation phase, the semantic package archiving module generates spatiotemporal semantic packages and writes spatiotemporal semantic package identifiers. The judgment conclusion, the set of business event atoms associated with the judgment conclusion, the sequence of evidence record entries associated with the set of business event atoms, the original perceived data reference information, process information, manual operation records, time information and location information, among which the spatiotemporal semantic package identifier. The millisecond-level timestamp string generated by the string prefix "PKG" and the judgment conclusion, as well as the minimum and maximum values ​​of the incrementing sequence numbers of all evidence record entries within the minimum evidence closure, are concatenated in a fixed order to ensure locatability and uniqueness.

[0105] To ensure that the packet hash digest calculation is recalculated and to avoid hash inconsistencies caused by differences in field order, the semantic packet archiving module organizes the spatiotemporal semantic packet content into a byte sequence to be hashed in a fixed order. The process involves writing the original perception data reference information segment first, then the evidence record entry sequence segment, then the process information segment, then the manual operation record segment, and finally the judgment conclusion segment. Each segment is serialized using standardized JSON with UTF-8 encoding and meets the following requirements: field names are sorted in lexicographical order, array elements are sorted in ascending order by incremental sequence number or event time, numerical values ​​are represented in decimal without exponent, and time fields are uniformly represented as millisecond-level timestamp strings. Segments are concatenated with each other using the format of "segment length decimal string + separator + segment byte stream" to eliminate concatenation ambiguity.

[0106] The semantic packet archiving module performs a process on the byte sequence to be hashed. Calculate packet hash digest And as a benchmark for integrity verification, the packet hash digest According to the formula:

[0107] ;

[0108] in The SHA represents the packet hash digest of the spatiotemporal semantic packet, with a length of 32 bytes. This refers to a cryptographic hash function identified by the SHA-256 algorithm. This represents the sequence of bytes to be hashed generated according to the fixed order and normalized encoding rules.

[0109] The semantic packet archiving module digests packet hashes. Submit to a third-party timestamp service and generate a timestamp token in accordance with RFC3161. And will As external proof of the formation time of the spatiotemporal semantic package, the package hash digest is also obtained using the private key of the archiving subject. Perform digital signature generation to generate signature value The certificate chain used for signature verification is also saved, with the digital signature algorithm fixed at the ECDSA curve. Furthermore, the hash algorithm is fixed to use SHA-256 to ensure that the signature is verifiable and the parameters are fixed;

[0110] Finally, the semantic packet archiving module combines the spatiotemporal semantic packets and packet hash digests. Timestamp token Digital signature The certificate chain is written to the archive storage as a single-set archive unit and is permanently saved using an append-only, non-overwrite write strategy, while a spatiotemporal semantic packet identifier is established. The index records are indexed using the primary key and the incrementing sequence number range of the judgment conclusion generation time field and the evidence record entry sequence as the retrieval fields to support subsequent fast retrieval by conclusion and the use of packet hash digests. Perform integrity checks and trace the formation process of hash associations between adjacent entries in the evidence record sequence.

[0111] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0112] This invention uses a combination of algorithms, namely "business event extraction and atomization, process diagram construction and representation learning, and single-set archiving unit generation based on minimum evidence closure", to unify the evidence chain organization and semantic expression of the original perceived data, process information, manual operation records and judgment conclusions to be preserved. Specifically, firstly, hash digests are calculated for each type of data, and evidence record entries are generated in chronological order. Adjacent entries are linked by hash to form a continuous sequence of evidence record entries, ensuring a continuous and verifiable link in the data's formation process from collection and transmission to processing and judgment. Next, key elements in the business process are extracted as business event atoms, and these atoms are bound to corresponding evidence record entry identifiers and hash digests, thus mapping business semantics to a verifiable chain of evidence. Then, a process diagram is used to explicitly model the temporal sequence, spatial relationships, processing dependencies, and relationships with manual operations. Based on this, the minimum evidence closure that satisfies the proof constraints is solved around the judgment conclusion. Finally, this is encapsulated into a spatiotemporal semantic package and stored using package-level hashing, timestamps, and digital signatures. This ensures that the conclusion is supported by the minimum necessary evidence set when issuing evidence, while guaranteeing the integrity, traceability, and non-repudiation of the evidence, thereby improving the ability to interpret, verify, and reproduce evidence, and increasing the efficiency of evidence issuance.

[0113] In terms of algorithm structure, this invention addresses the technical problems of "weak cross-stage correlation, difficulty in process verification, and difficulty in single-set expression and review of archives" by making evidence-oriented improvements: First, it incorporates evidence record entry nodes and business event atomic nodes into a heterogeneous process graph, and uses relationship type fields to distinguish temporal sequence, spatial correlation, processing dependency, and manual operation relationships, enabling human-machine collaboration traces to be expressed in a structured manner and subject to constraints and verification. Second, it uses process graph representation learning to generate graph representation features for traceability search, and introduces an evidence contribution-driven elimination and rollback mechanism. After each elimination, it verifies the proof constraints such as temporal continuity, processing dependency integrity, and traceability of manual operations, ensuring that the closure is both "as small as possible" and "provable" in the solution process. Third, by incorporating processing program identification information, version information, processing parameter information, and runtime environment fingerprint information into the proof constraints and archive content, it enhances the reproducibility and verification consistency of the processing process. These improvements enable the case to form a more stable continuous chain of evidence and generate a verifiable, reviewable, and reproducible single-set spatiotemporal semantic package centered on the conclusion.

Claims

1. A method for generating a continuous human-machine collaboration spatio-temporal semantic package based on evidence chains, characterized in that, include: S1. Collect spatiotemporal data corresponding to the target business process, and synchronously record the process information of spatiotemporal data in the process of collection, transmission, processing and judgment to form an evidence dataset. Generate hash digests for the original perception data, process information, manual operation records and judgment conclusions in the evidence dataset, generate evidence record entries in chronological order, and form a continuous linked sequence of evidence record entries based on the hash association between adjacent evidence record entries. S2. Based on the evidence dataset and evidence record item sequence, perform business event extraction and atomic processing on the original perception data, process information, manual operation records and judgment conclusions to generate a business event atom set, and associate the corresponding evidence record item identifier and its hash digest with each business event atom; S3. Construct a process diagram representing the human-machine collaboration process based on the business event atomic set and the evidence record item sequence, including business event atomic nodes and evidence record item nodes, as well as edges representing temporal sequence, spatial correlation, processing dependency and manual operation relationship; S4. Perform process graph representation learning on the process graph to generate a graph representation feature set. Perform traceability solution on the atomic nodes of the business events corresponding to the judgment conclusion. Under the condition of satisfying the preset proof constraints, select the minimum number of nodes and edges required to support the judgment conclusion from the process graph to generate the minimum evidence closure. The preset proof constraints include at least time continuity constraints, processing dependency integrity constraints, and manual operation traceability constraints. S5. Encapsulate the minimum evidence closure to generate a spatiotemporal semantic package, and archive the spatiotemporal semantic package as a single-set archiving unit. Generate a package hash digest for the spatiotemporal semantic package, generate a timestamp for the package hash digest, and digitally sign and store it to support the integrity verification of the spatiotemporal semantic package and the traceability of its formation process.

2. The method for generating spatiotemporal semantic packets based on continuous evidence chains in human-computer collaboration according to claim 1, characterized in that, S1 includes: Collect raw sensing data corresponding to the target business process, and simultaneously generate time information and location information corresponding to the raw sensing data during the collection process; The process information of the original sensing data during the acquisition, transmission, processing and judgment process is recorded synchronously. The process information includes at least the acquisition device identification information, acquisition parameter information, transmission link identification information, processing program identification information, processing parameter information and operator identification information. The original sensing data, the process information, the manual operation record, and the judgment conclusion are standardized and encoded respectively, and then hash digests are calculated. An incremental sequence number is assigned to the hash digest corresponding to each item. Evidence record entries are generated in the order of the incrementing sequence number. In each evidence record entry, the incrementing sequence number of the evidence record entry, the hash digest of the evidence record entry, the time information corresponding to the evidence record entry, the location information corresponding to the evidence record entry, and the hash digest of the previous evidence record entry are written. Each evidence record entry is appended and stored in the order of its generation to form the continuously linked evidence record entry sequence, and the original perception data, the process information, the manual operation records, and the judgment conclusions, together with their corresponding hash digests, are collected to form the evidence dataset.

3. The method for generating spatiotemporal semantic packets based on continuous evidence chains in human-computer collaboration according to claim 1, characterized in that, S2 include: Based on the time and location information carried in the evidence dataset, the original perception data, process information, manual operation records and judgment conclusions are time-aligned and spatially aligned to form candidate event fragments arranged in chronological order. For each candidate event fragment, perform event element extraction to obtain the event time, event location, event subject, event action, and event object corresponding to the candidate event fragment; The event elements are atomically segmented according to the preset event template to generate business event atoms, and a unique event identifier is assigned to each business event atom. Based on the event time and event location of each business event atom, a matching evidence record entry is retrieved in the evidence record entry sequence. The incrementing sequence number of the retrieved evidence record entry is determined as the evidence record entry identifier of the business event atom, and the hash digest of the evidence record entry is determined as the hash digest of the business event atom. All business event atoms are aggregated to form a business event atom set, where each business event atom includes at least an event identifier, event time, event location, event subject, event action, event object, corresponding evidence record entry identifier, and corresponding hash digest.

4. The method for generating spatiotemporal semantic packets based on continuous evidence chains in human-computer collaboration according to claim 1, characterized in that, S3 includes: Each business event atom in the business event atom set is constructed as a business event atom node, and each evidence record entry in the evidence record entry sequence is constructed as an evidence record entry node, thus forming a node set of the process diagram; The business event atomic nodes are sorted according to the event time in the business event atomic set, and time sequence edges are established between adjacent business event atomic nodes in the sort. Based on the event location in the business event atomic set, establish spatial association edges for business event atomic nodes that meet the spatial adjacency conditions according to the preset spatial adjacency rules; Based on the process information in the evidence dataset, identify the calling relationship or data dependency relationship between the data processing records corresponding to the atomic business events, and establish processing dependency relationship edges for the atomic nodes of business events that have calling relationships or data dependencies. Based on the manual operation records in the evidence dataset, identify the association between the manual operation records and the corresponding business event atoms, and establish a manual operation relationship edge between the evidence record entry node corresponding to the manual operation record and the corresponding business event atom node. Write the relationship type field for the time sequence relationship edge, spatial relationship relationship edge, processing dependency relationship edge and manual operation relationship edge respectively, and write the edge weight for each edge according to the preset weighting rule to obtain the process diagram representing the human-computer collaboration process.

5. The method for generating spatiotemporal semantic packets based on continuous evidence chains in human-computer collaboration according to claim 1, characterized in that, S4 includes: Write node attributes for the business event atomic nodes in the process diagram. The node attributes include at least event time, event location, event subject, event action, event object, evidence record entry identifier, and hash digest. Write node attributes for the evidence record entry nodes in the process diagram. The node attributes include at least incrementing sequence number, time information, location information, and hash digest. Input the node attributes and the relationship type field and edge weight of each edge in the process graph into the process graph representation learning model to obtain the graph representation feature set corresponding to each node and each edge; Based on the feature set of the graph representation, the evidence contribution degree of the atomic node of the business event corresponding to the judgment conclusion is calculated, and the evidence contribution degree is used as the search weight to perform a retrospective search in the process graph to obtain a candidate closed subgraph containing the atomic node of the business event corresponding to the judgment conclusion. In the candidate closure subgraph, nodes and edges are removed in order of their evidence contribution from low to high. After each removal, the preset proof constraints are verified. If the verification passes, the removal result is retained. If the verification fails, the removal result is rolled back until there are no nodes or edges in the candidate closure subgraph that can be removed and still satisfy the preset proof constraints, thus obtaining the minimum evidence closure. The preset proof constraints include at least a time continuity constraint, a processing dependency integrity constraint, and a manual operation traceability constraint. The time continuity constraint limits the atomic nodes of the business events corresponding to the judgment conclusion in the minimal evidence closure to be traceable along the time sequence edge to the preceding atomic nodes of the business events with corresponding evidence record entries. The processing dependency integrity constraint limits the minimal evidence closure to include all preceding atomic nodes of the business events corresponding to the judgment conclusion that have processing dependency edges. The manual operation traceability constraint limits the minimal evidence closure to include evidence record entries that have manual operation edges of the business events corresponding to the judgment conclusion.

6. The method for generating spatiotemporal semantic packets based on continuous evidence chains in human-computer collaboration according to claim 1, characterized in that, S5 include: Based on the minimum evidence closure, the original perceived data reference information, process information, manual operation records and judgment conclusions corresponding to the business event atomic nodes and evidence record entry nodes contained in the minimum evidence closure are extracted and summarized to obtain the spatiotemporal semantic package content. The content of the spatiotemporal semantic package is structured and encapsulated to generate a spatiotemporal semantic package. The spatiotemporal semantic package includes at least a spatiotemporal semantic package identifier, a judgment conclusion, a set of business event atoms associated with the judgment conclusion, a sequence of evidence record entries associated with the set of business event atoms, original perception data reference information, process information, manual operation records, time information, and location information. The original sensory data reference information, the evidence record entry sequence, the process information, the manual operation record, and the judgment conclusion in the spatiotemporal semantic package are concatenated in a preset order and a packet hash digest is calculated. The packet hash digest is generated with a timestamp, digitally signed, and then stored. The spatiotemporal semantic package, the package hash digest, the timestamp, and the digital signature are written into the archive storage as a single-set archive unit to support the integrity verification of the spatiotemporal semantic package and the traceability of its formation process.

7. The method for generating spatiotemporal semantic packets based on continuous evidence chains in human-computer collaboration according to claim 2, characterized in that, The hash digests corresponding to the evidence record entries generated within the preset time window are constructed into a Merkle tree, and the Merkle root hash is sent as the anchor digest to a third-party timestamp service or distributed ledger for evidence storage. The returned evidence storage receipt is written into the sequence of evidence record entries.

8. The method for generating spatiotemporal semantic packets based on continuous evidence chains in human-computer collaboration according to claim 5, characterized in that, The process graph representation learning model is a heterogeneous relational graph neural network model. Different relational parameters are used for message passing on the temporal sequence relational edges, spatial relational edges, processing dependency relational edges, and manual operation relational edges. Time position encoding is introduced for event time to generate the graph representation feature set.

9. The method for generating spatiotemporal semantic packets based on continuous evidence chains in human-computer collaboration according to claim 5, characterized in that, The preset proof constraint further includes a process reproducibility constraint, which is used to limit the minimum evidence closure to include at least the process identification information, process version information, process parameter information, and runtime environment fingerprint information related to the judgment conclusion; wherein, the runtime environment fingerprint information includes at least one of container image digest, runtime library version digest, or execution environment proof digest.

10. A human-computer collaborative spatiotemporal semantic packet generation system based on a continuous evidence chain, used to execute the human-computer collaborative spatiotemporal semantic packet generation method based on a continuous evidence chain as described in any one of claims 1 to 9, characterized in that, include: The evidence chain generation module is used to collect spatiotemporal data of the target business process and record the corresponding process information. It generates hash digests of the original perception data, process information, manual operation records and judgment conclusions, generates evidence record entries in chronological order and forms a continuous evidence record entry sequence by hash association between adjacent entries. The event atomization module is used to extract and atomize business events based on the evidence data and the sequence of evidence record entries, generate a set of business event atoms and associate each business event atom with the corresponding evidence record entry identifier and its hash digest. The process diagram construction module is used to construct process diagrams using business event atomic nodes and evidence record entry nodes, and to establish edges for temporal sequence, spatial association, processing dependency, and manual operation relationships. The minimum evidence closure generation module is used to perform representation learning on the process graph and trace back to solve the atomic nodes of the business events corresponding to the judgment conclusion. Under the conditions of satisfying the constraints of time continuity, processing dependency integrity and traceability of manual operation, the minimum evidence closure supporting the judgment conclusion is generated. The semantic package archiving module is used to encapsulate the minimum evidence closure into a spatiotemporal semantic package and archive it as a single-set archiving unit. It also generates a package hash digest for the spatiotemporal semantic package and stores it with a timestamp and digital signature to support integrity verification and traceability of the formation process.

Images